Multi-fidelity active learning with GFlowNets for drug and materials discovery.

14^th September 2023, 14:00
Alex Hernandez-Garcia
Mila AI Quebec Institute, Canada

Abstract

In recent years, machine learning has been increasingly adopted by researchers from different disciplines as a tool to accelerate the pace to scientific discoveries. However, many relevant scientific problems present challenges where current machine learning methods cannot yet efficiently leverage the available data and resources. For example, such tasks involve exploring very large, high-dimensional spaces, where querying a high fidelity, black-box objective function is very expensive. Progress in machine learning methods that can efficiently tackle such problems would help accelerate currently crucial areas such as drug and materials discovery. In this talk, I will present our work on the use of generative flow networks (GFlowNets) for multi-fidelity active learning, where multiple approximations of the black-box function are available at lower fidelity and cost. GFlowNets are a recently proposed framework for amortised probabilistic inference that have proven efficient for exploring large, high-dimensional spaces and can hence be practical in the multi-fidelity setting too. I will provide a gentle introduction to GFlowNets, describe our algorithm for multi-fidelity active learning with GFlowNets and present the results on both well-studied synthetic tasks and practically relevant applications of molecular discovery. Finally, I will discuss future directions in materials discovery applications where we plan to use multi-fidelity active learning with GFlowNets.